Re: [PATCH] driver core: platform: fix race condition with driver_override

[Greg KH]

On Wed, Apr 26, 2017 at 07:51:32AM -0700, Adrian Salido wrote:
> > > The driver_override implementation is susceptible to race condition when
> > > different threads are reading vs storing a different driver override.
> > > Add locking to avoid race condition.
> > >
> > > Fixes: 3d713e0e382e ("driver core: platform: add device binding path 'driver_override'")
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Signed-off-by: Adrian Salido <salidoa@xxxxxxxxxx>
> > > ---
> > >  drivers/base/platform.c | 11 +++++++++--
> > >  1 file changed, 9 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> > > index c2456839214a..493e03fa0e07 100644
> > > --- a/drivers/base/platform.c
> > > +++ b/drivers/base/platform.c
> > > @@ -866,7 +866,7 @@ static ssize_t driver_override_store(struct device *dev,
> > >                                    const char *buf, size_t count)
> > >  {
> > >       struct platform_device *pdev = to_platform_device(dev);
> > > -     char *driver_override, *old = pdev->driver_override, *cp;
> > > +     char *driver_override, *old, *cp;
> > >
> > >       if (count > PATH_MAX)
> > >               return -EINVAL;
> > > @@ -879,12 +879,15 @@ static ssize_t driver_override_store(struct device *dev,
> > >       if (cp)
> > >               *cp = '\0';
> > >
> > > +     device_lock(dev);
> > > +     old = pdev->driver_override;
> > >       if (strlen(driver_override)) {
> > >               pdev->driver_override = driver_override;
> > >       } else {
> > >               kfree(driver_override);
> > >               pdev->driver_override = NULL;
> > >       }
> > > +     device_unlock(dev);
> > >
> > >       kfree(old);
> >
> > Shouldn't you move the lock until after the kfree()?  Or am I missing
> > what the lock is trying to protect here?
> 
> not really, the lock only protecting the variable
> pdev->driver_override. Once the value has changed we no longer care
> about "old" variable

What are you protecting it from?  Being overwritten twice?  Or something
else?

> > >       if (cp)
> > >               *cp = '\0';
> > >
> > > +     device_lock(dev);
> > > +     old = pdev->driver_override;
> > >       if (strlen(driver_override)) {
> > >               pdev->driver_override = driver_override;
> > >       } else {
> > >               kfree(driver_override);
> > >               pdev->driver_override = NULL;
> > >       }
> > > +     device_unlock(dev);
> > >
> > >       kfree(old);
> >
> > >
> > > @@ -895,8 +898,12 @@ static ssize_t driver_override_show(struct device *dev,
> > >                                   struct device_attribute *attr, char *buf)
> > >  {
> > >       struct platform_device *pdev = to_platform_device(dev);
> > > +     ssize_t len;
> > >
> > > -     return sprintf(buf, "%s\n", pdev->driver_override);
> > > +     device_lock(dev);
> > > +     len = sprintf(buf, "%s\n", pdev->driver_override);
> > > +     device_unlock(dev);
> > > +     return len;
> >
> > Why does the show function need to be changed at all?  How can anything
> > "race" here?
> 
> The lock is trying to protect again race between store and show.
> Suppose there are 2 threads:
> 
> Thread1:
>  while (1) {
>    driver_override_store("foo");
>    driver_override_store("");
>  }
> 
> Thread2:
> while (1) driver_override_show();
> 
> Thread 1                                 |         Thread 2
> ----------------------------------------|-----------------------
> old = pdev->driver_override;   |
>                                                 | len = sprintf(buf,
> "%s\n", pdev->driver_override);
>                                                 | /* snprintf starts reading */
> pdev->driver_override =          |
>    driver_override;                    |
> kfree(old);                                | /* use after free before
> snprintf finishes execution */
> 
> Similarly there could be a race between multiple threads doing store
> where memory leaks could happen

Ah, the printing of the string, that makes more sense, thanks, I was
thinking of the assignment of the pointer itself, which is atomic on all
sane platforms.

As writing this is only allowed by root, it's not really a big deal,
I'll queue it up for the next release.

thanks,

greg k-h